Sampo Pyysalo

Sampo Pyysalo is a natural language processing researcher at the University of Turku whose work runs from biomedical text mining to open Finnish and European language models. After a PhD thesis on dependency parsing for biomedical text mining at Turku in 2008, he co-authored the BioNLP'09 shared-task overview in 2009 and the brat annotation tool in 2012, and held researcher posts at the University of Tokyo, the University of Manchester and the University of Cambridge. Back in Turku's TurkuNLP group from 2019, he co-authored a Finnish BERT model in 2019, the FinGPT models in 2023 and the 34-billion-parameter Poro model in 2024, along with Scaling Data-Constrained Language Models, a NeurIPS 2023 Outstanding Main Track Runner-Up. He is now a Research Fellow in the Department of Computing at Turku, co-leads TurkuNLP, and leads the work package on creating LLMs in the OpenEuroLLM project.

Now at: University of Turku
Works: 298
Citations: 29,129
h-index: 47
Badges: Boomerang

Publication numbers from OpenAlex, as of 22 October 2026.

Timeline

2008

A PhD at Turku

He defends a PhD thesis in computer science at the University of Turku. The thesis, "A Dependency Parsing Approach to Biomedical Text Mining", appears in 2008 as TUCS Dissertations 105 from the Turku Centre for Computer Science.
2009–2012

The BioNLP'09 shared task and brat

In 2009, he co-authors the overview of the BioNLP'09 Shared Task on Event Extraction with Jin-Dong Kim, Tomoko Ohta, Yoshinobu Kano and Jun'ichi Tsujii. In 2012, he co-authors brat, a web-based tool for NLP-assisted text annotation, with Pontus Stenetorp, Goran Topić, Tomoko Ohta, Sophia Ananiadou and Tsujii.
2019

A Finnish BERT and a return to Turku

After researcher posts at the University of Tokyo, the University of Manchester and the University of Cambridge, including work at the UK's National Centre for Text Mining, he returns to the TurkuNLP group at Turku in 2019. In December 2019, he and seven coauthors introduce a Finnish BERT model trained from scratch that outperforms multilingual BERT and sets new state-of-the-art results in part-of-speech tagging, named entity recognition and dependency parsing.
2020–2021

Finnish named entity corpora

In May 2020, he co-authors the Turku NER corpus at LREC, which adds name and date annotation to the 754-document Universal Dependencies Finnish corpus across ten genres, with over 10,000 mentions. In 2021, at the Nordic Conference on Computational Linguistics (NoDaLiDa), he co-authors "Fine-grained Named Entity Annotation for Finnish", a corpus of nearly 500,000 tokens with over 50,000 mentions in the 18 OntoNotes entity types. These manual annotation efforts produce the Turku NER and TurkuONE corpora.
May 2023–December 2023

A NeurIPS runner-up paper

In May 2023, he is one of nine authors of "Scaling Data-Constrained Language Models", which trains models of up to 9 billion parameters on up to 900 billion tokens in 400 training runs and finds that up to 4 epochs of repeated data change loss very little compared with unique data. NeurIPS names the paper an Outstanding Main Track Runner-Up in December 2023.
August 2023–April 2024

An open LLM consortium and Poro 34B

In August 2023, the TurkuNLP group and SiloGen, the large language model arm of Silo AI, launch a consortium to build a family of open LLMs with access to the LUMI supercomputer; Pyysalo, a University Research Fellow and principal investigator of the High Performance Language Technologies consortium at Turku, is on the team. In April 2024, he co-authors the paper introducing Poro 34B, a 34-billion-parameter model trained on 1 trillion tokens of Finnish, English and programming languages.
November 2023

FinGPT for a small language

In November 2023, he and 20 coauthors present FinGPT, seven monolingual Finnish models from 186 million to 13 billion parameters trained from scratch. The work also releases BLUUMI, a 176-billion-parameter model built by continuing the pretraining of the multilingual BLOOM model with Finnish, and FIN-bench, a Finnish version of BIG-bench.
As of October 2026

TurkuNLP and OpenEuroLLM

He is a Research Fellow in the Department of Computing at the University of Turku and a co-lead of the TurkuNLP group, whose Finnish models include FinBERT, FinGPT, Poro and the Viking family. He leads the work package on creating LLMs in the OpenEuroLLM project, which builds fully open foundation models for all European languages. As its technical lead, he welcomes over 10 million GPU hours of strategic EuroHPC compute for the project in December 2025.